This fixes the cases where the yaw message from the GNSS receiver would
take more time than the vel/pos. The estimator should wait and not
immediately fall back to an other aiding source after 5sec.
If it never comes, it will never fall back, but this is ok since the
user wants to fly with GPS aiding an not with something else.
If the user selects GPS yaw fusion but that there is no GPS yaw data in
the GPS message or if the fusion is rejected for some time, the GPS yaw
data is declared faulty and the fusion is stopped to allow an other
source of yaw aiding to start.
- do not gps reference altitude to zero in case gps checks pass before the
filter initialized
- reset the filtered gps position and position derivative filters in case
we are in air or there is movement on the ground
Signed-off-by: RomanBapst <bapstroman@gmail.com>
This is how it is also done in ekf2_main. Otherwise, this leads to
multiple integration of the same IMU data due to asynchronous sensor
updates triggering a prediction step between IMU updates.
Fix unit tests that broke because of this fix
The `_deadreckon_time_exceeded` flag is used in
`local_position_is_valid()`. This means that
`_params.valid_timeout_max` after startup, in my observed case 5
seconds, the local position switche from valid to invalid and then after
a while back to valid again.
With this fix, the local position is flagged invalid from boot and gets validated after the first aiding event.
Co-authored-by: Julian Oes <julian@oes.ch>
* ekf: disable xy accel bias learning before takeoff
As those biases are usually poorly observable before takeoff because
they are almost perpendicular to the gravity vector, learning is often
driven by noise and numerical issues. This results in incorrect bias
learning before takeoff when the drone is static on ground for a long
period of time.
* ekf: update unit test and change indicator
This is a non-functional change required to select accel bias estimation
per axis selection. The intent is then to disable the learning before
takeoff of the components that are poorly observable.
kamilritz
Daniel Agar
sevenbill
Bill Morris
PX4 BuildBot
bresch
Matthias Grob
RomanBapst
Paul Riseborough
Paul Riseborough
Carl Olsson